Re: VSTa update

From: David Jeske <jeske_at_nospam.org>
Date: Sat Mar 11 2000 - 21:37:05 PST

I don't mean for this to degenerate into a "Linux graphics"
conversation on the VSTa list.. however, there are some relevant
pieces here.

On Sat, Mar 11, 2000 at 02:28:54PM -0600, Major wrote:
> Anyway .. the linux kernel FB is an abstract interface to a graphics adapter
> that allows for the userland application to supply it's own acceleration.
> Thusly a new svgalib, or libggi, or X11 need supply the graphical acceleration
> for the FB device. Linux FB is not a textmode thing at all.

I see. I was thinking of the older GGI stuff which had more
complicated graphics drivers (fb + accel) in the kernel before
Linux-/dev/fb existed.

I'm very opposed to the "library which has hardware drivers in it"
model, since any code which touches hardware directly can crash the
machine pretty trivially. (I'm sure I can cause at least two or three
PCI graphics cards to lock up the PCI bus completely with just a few
MMIO writes.)

IMO, the nicest part about the previous GGI (the part I presume you
were calling bloat), is that it put the code which needed to be
trusted (all the code which touched the graphics hardware) into a
"trusted space", which in that case was the kernel.

What they've done now (IMO) basically requires everything to go
through X, so that X can be the trusted userland server with
acceleration support. I like the idea of this code living in a trusted
userspace graphics server (instead of kernel), but X is relatively
annoying to configure and is not as modular as it should be. (i.e. you
can't just install a binary graphics driver, and XF86Config is a
pain). In addition, until Xfree4/DRI, there wasn't any way to get
low-level access to hardware acceleration through X. I don't even know
how well that stuff works, since it's only recently been released.

In addition, with Xfree4/DRI, from what I understand, they are
planning for the OpenGL libraries to directly touch the graphics
hardware from your application. SGI spent years designing hardware
which was "safe" enough to touch from userspace, and which had
hardware virtualization. In the PC world, hardware will never be like
this. The Windows-esq DirectX model is (IMO) better as it allows
trusted drivers, untrusted applications, cheap software sharing (but
not pure virtualization), and relatively fast access to the raw
hardware acceleration features.

-- 
David Jeske (N9LCA) + http://www.chat.net/~jeske/ + jeske_at_chat.net
Received on Sat Mar 11 21:44:38 2000

This archive was generated by hypermail 2.1.8 : Thu Sep 22 2005 - 15:12:56 PDT